The DiscoSNC currently offers NumberObj and NumberObjType, a class and an enum, to switch the tool's number backend. This functionality does not seem to be used, see comment FIXME Implement a global flag in Analysis.java (?) and read that out. in line 28 of NumberObj. I assume the provided code is thus not tested either (?).

The DiscoDNC offers a switchable number backend, too. This backend provides wrappers for primitive single and double precision floating point numbers as well as rationals based on Integer or BigInteger (with additional classes to add +/-infinity and NaN to those). Since commit NetCal/DNC@88dd059 it is entirely independent of the DiscoDNC. It can be packaged separately and used as a library (the plan is to move it into its own project eventually). Moreover, this number backend is tested by the DiscoDNC functional tests.

Thus, I suggest to remove NumberObj and NumberObjType in favor of code sharing with the DiscoDNC. Currently, this task does not seem to cause a big porting effort.
Comments like the one above should be added to the locations in the code that require adaptation for the switchable number backend.

Commands' constructors in de.uni_kl.cs.disco.snc.calculator.commands take the networkID and store them in a member variable. However, when calling execute(), the SNC's current network is queried with snc.getCurrentNetwork().

An exception is EndToEndConvolutionCommand. It gets the current network in its constructor and stores it in a member variable. It doesn't even store the given network ID at all and in turn the stored SNC variable is never used.

Neither the snc.jar to download nor a newly build one starts on my system (macOS 10.13.6, Oracle Java 10). The Java icon appears for a short time but disappears again. No GUI is shown, as if the application crashed.

Further testing shows:

• Using a Java 8 JRE, I can start the snc.jar that can be downloaded. But the left panel is too small and cannot be enlarged.
• Compiling with Maven compiler.source and compiler.target 1.8 does not start with a JRE 8.

in the method public double getValue(double theta, Map<Integer, Hoelder> parameters) there are checks for original.getParameters().containsKey(hoelder).

original is an object of SymbolicFunction and getParameters() returns Map<Integer, Hoelder>, i.e., keys are Integer objects.

Should the code test containsValue(hoelder)?

In SimpleOptimizer.java, method public double minimize(...) there is this loop:

while(theta < maxTheta) {
	try {
		Math.min(optValue, bound.evaluate(theta));
		theta = theta+thetagranularity;
	} catch(ServerOverloadException | ThetaOutOfBoundException e) {
		theta = theta+thetagranularity;
	}
}

and Spotbugs rightfully complains that the result of Math.min is not used.
Note, that bound.evaluate(theta) cannot change theta as it is an immutable double.

Moreover, above this problematic loop, we find similar loops. For example:

while(theta < maxTheta) {
	try {
		optValue = Math.min(optValue, bound.evaluate(theta));
		theta = theta+thetagranularity;
	} catch(ServerOverloadException | ThetaOutOfBoundException e) {
		theta = theta+thetagranularity;
	}
}

They all assign optValue = . Is that assignment missing in the problematic loop?